Conference PaperPDF Available

Analysis of satellite monthly precipitation time series over East Africa

Authors:

Abstract and Figures

East Africa experienced in the 2001 -2011 time period some of the worst drought events to date, culminated with the high-impact drought in 2010-2011. The frequency and impacts of these extreme events require a continuous monitoring of precipitation, as a key variable for the inclusion of these phenomena in regional climatological studies and their timely forecast. Satellite precipitation products are particularly necessary in the region to enhance the observational capabilities limited sparse rain-gauge networks. Nevertheless, East Africa is characterized by a complex topography and highly varying climatic conditions ranging from the wetter mountainous regions to the arid lowlands with different precipitation seasonality, which can greatly affect the quality of satellite rainfall estimations. It is thus of utmost importance a satellite product validation and inter-comparison in order to assess their reliability and delimit the application domain. The monthly accumulated precipitation from seven satellite products, TAMSAT, GSMaP, CMORPH, PERSIANN, RFE, TRMM-3B42, and TRMM-3B31 are analysed for the time period 2001-2009, by dividing the studied region (5°S–20°N, 28°E–52°E) in eight sub-areas (clusters) characterized by a different annual cycle. Clusters are identified by applying a non-hierarchical k-mean cluster analysis to the GPCC Climatology Version 2011 product. The variability of the satellite-based precipitation estimations is firstly evaluated by computing the variance from the ensemble of the seven satellite products, after projecting them on a common latitude and longitude grid. The variability among satellite products shows a dependence on season, precipitation intensity, and topography. Further comparisons (correlation coefficient, mean error, root mean square error, and efficiency coefficient) are carried out with the GPCC Full Data Reanalysis at 0.5° resolution, taking into account the characteristics of the different clusters. Finally, monthly anomalies between the satellite products and the GPCC Climatology Version 2011 product are computed to evaluate the potential of satellite products for identifying the drought periods.
No caption available
… 
No caption available
… 
No caption available
… 
Content may be subject to copyright.
Analysis of satellite monthly precipitation time series over East Africa
E. Cattani1, A. Merino2, C. Wenhaji Ndomeni1, and V. Levizzani1
1CNR-ISAC, Bologna, Italy
2Group for Atmospheric Physics, IMA, University of León, León, Spain
7th International Precipitation Working Group Workshop, 17 21 November 2014
Tsukuba International Congress Center, Tsukuba, Japan
Annual precipitation from GPCC Climatology at 0.5°
Precipitation variability over East Africa (1/3)
East Africa (EA) is characterized by
complex topography and highly varying
climatic conditions, which reflect into a
marked geographic variability of
precipitation...
(mm/year)
Precipitation variability over East Africa (2/3)
... clustering…
Eight areas (clusters) were identified by applying a
non-hierarchical k-mean cluster analysis to the
GPCC_CLIM data on the basis of the characteristics
of the precipitation annual cycle.
Meyer-Christoffer, A. et al. GPCC Climatology Version 2011 at 0.25°: Monthly Land-Surface Precipitation Climatology for Every Month and the Total Year from Rain-Gauges
built on GTS-based and Historic Data. doi: 10.5676/DWD_GPCC/CLIM_M_V2011_025 (2011).
Clusters
1, 3, 6 & 8 Southeastern Ethiopia, Somalia,
Kenya, Southern Uganda and Tanzania
2 Central Ethiopia (Rift Valley, W & E
Highlands escarpments), part of Uganda,
Kenya
4 Mountainous coastal Somalia (Gulf of
Aden), Central Uganda, Congo
5 Sudan, Western Ethiopian Highlands
7 Northern Sudan, coastal Eritrea
Precipitation variability over East Africa (3/3)
... and different precipitation seasonality.
Mean annual cycles from GPCC Climatology
The complexity of precipitation seasonality
is due to the superimposition of large-scale
climatic controls (ITCZ), and regional factors
(lakes, topography, etc.).
Meyer-Christoffer, A. et al. GPCC Climatology Version 2011 at 0.25°: Monthly Land-Surface Precipitation Climatology for Every Month and the Total Year from Rain-Gauges
built on GTS-based and Historic Data. doi: 10.5676/DWD_GPCC/CLIM_M_V2011_025 (2011).
Drought events and precipitation measurements
Produced by EUMETSAT in collaboration with the COMET® Program
The frequency and impacts of these extreme
events require a continuous monitoring of
precipitation as a key variable for the inclusion
of these phenomena in regional climatological
studies and their timely forecast.
Satellite precipitation products are particularly
necessary in EA to enhance the observational
capabilities of sparse rain-gauge networks.
Precipitation measurements from satellite
Algorithm Spatial
resolution Time
resolution Input data Producer/reference
3B42_daily v7 0.25° daily MW, GEO-IR, gauges NASA/GSFC
Huffman et al., 2007
CMORPH v1
bias corrected 0.25° daily MW, GEO-IR motion vectors,
daily gauges NOAA/CPC
Joyce et al., 2004
GSMaP_MVK v5.222.1 0.1° hourly MW, GEO-IR,
GEO-IR motion vectors EORC/JAXA
Aonashi et al., 2009
Kubota et al., 2007
Ushio et al., 2009
TAMSAT TARCAT v2
0.0375° monthly GEO-IR, climatological calibration
with rain gauges University of Reading
Grimes et al., 1999
Maidment et al., 2014
PERSIANN 0.25° 6-hourly GEO-IR, PMW rainfall estimates to
update model parameters UC Irvine
Hsu et al., 1997
3B31 v7 0.5° monthly TRMM TMI and PR NASA/GSFC, JAXA
Haddad et al., 1997
RFE v2
0.1° daily PMW, GEO-IR with GPI, GTS rain
gauges NOAA/CPC
Xie and Arkin, 1996
Monthly accumulated precipitation was calculated for each satellite product and for the time
period 2001-2009. All data sets were re-projected on a common grid at 0.5°.
Mean precipitation annual cycles (2001-2009) from satellite products at 0.5°
Satellite products reproduce the
characteristics of the precipitation cycle of
each cluster: wet season duration,
identification of peak intensity months, and
presence of a prevailing wet season in case
of bi-modal cycles.
cluster 1 cluster 2
cluster 3 cluster 4
cluster 5 cluster 6
cluster 8 cluster 7
TAMSAT
GSMaP
3B42
PERSIANN
RFE
CMORPH
3B31
GPCC_CLIM
Precipitation intensity distribution of satellite products at 0.5°
Consistency among the intensity precipitation distributions of the various satellite products.
All bi-modal clusters exhibit precipitation distributions with 2 peaks at 25-50 and 100-150 mm; the mono-modal cluster 5
has a occurrence peak in the range 100-300 mm, whereas in cluster 7 (arid region with very scarce precipitation) most of
the occurrences concentrate in the first 4 bins (< 50 mm).
Satellite products generally overestimate the occurrences with monthly precipitation < 10 mm with respect to GPCC_FD.
Schneider, U. et al. GPCC Full Data Reanalysis Version 6.0 at 0.5°: Monthly Land-Surface Precipitation from Rain-Gauges built on GTS-based and Historic Data.
doi:10.5676/DWD_GPCC/FD_M_V6_050 (2011)
TAMSAT
GSMAP
3B42
PERSIAN
N
RFE
CMORPH
3B31
GPCC_FD
Satellite precipitation estimate variability (1/2)
Seasonal six-member ensemble
means (2001-2009)
A first assessment of the differences
among the satellite products was
obtained by analysing the standard
deviations from the ensemble of the six
monthly satellite products (Tian and
Peters-Lidard, 2010).
This method is instrumental to identify
situations that reveal particularly
problematic for the precipitation
retrieval from satellite.
(mm)
Satellite precipitation estimate variability (2/2)
Mount Kenya (5199 m),
, 37° E
(mm)
Seasonal average standard deviations
from the six-member ensemble (2001-
2009)
Ensemble standard deviation is linked to:
precipitation intensity
precipitation seasonality
orography
Standard deviation even greater than
80 mm over West Ethiopian Highlands
in summer.
Comparisons between satellite products and GPCC_FD v6 at 0.5° (1/2)
Key points to be considered:
GPCC_FD is the most accurate in situ precipitation data set of GPCC, nevertheless the local
poor density (Somalia, Eritrea, and Djibouti) of rain stations has to be considered.
This affects the quality of the GPCC_FD and of the satellite products that make use of rain
gauge measurements.
Satellite products exploiting rain gauge measurements (3B42 in particular) are expected to
better agree with GPCC_FD data set. For these products comparisons with GPCC can not
be interpreted as a completely independent validation.
Statistical parameters:
EFF skill score of satellite estimates accuracy vs GPCC_FD
1 best score
0 satellite estimates are as accurate as the rain gauge
mean value
<0 rain gauge mean value is a better estimate than
satellite estimates
Comparisons between satellite products and GPCC_FD v6 at 0.5° (2/2)
1. TRMM-3B42 best performance satellite product in all clusters with MAE values in the range 5-25
mm/month and best Efficiency Score (EFF≈0.7 for all clusters) values.
2. RFE and TAMSAT exhibit similar performances in terms of MAE, RMSE, and EFF (>≈0.5), with lower BIAS
values for TAMSAT.
3. GSMaP results similar to those of CMORPH (MAE, RMSE, and EFF) with differences for cluster 5.
GSMaP’s BIAS values are more variable.
4. PERSIANN and 3B31 show very low EFF, even negative.
Comparisons between satellite products and GPCC_FD v6: the terrain elevation (1/3)
The complex orography can be an issue for the precipitation retrieval from satellite:
IR-based retrievals can have problems to identify warm orographic rainfall
MW-based retrievals rely on ice scattering over land, which can be moderate in case of warm orographic rain
The presence of snow or ice on the ground is a further difficulty for the MW-based retrieval
Monthly mean precipitation as a function of elevation.
Generally precipitation intensity increases with elevation (H > 250-500 m) with a trend which depends on clusters.
Satellite products give similar results over grid cells with H < 250-500 m.
Larger differences among satellite products are detected for H > 250-500 m.
Comparisons between satellite products and GPCC_FD v6: the terrain elevation (2/3)
Monthly mean precipitation as a function of elevation.
For H > 1000-1500 m the precipitation increase with elevation becomes weaker (clusters 1, 2, 3, 4, 5, 8).
Over high-elevated grid cells 3B31, 3B42, and CMORPH exhibit the highest precipitation intensity values.
Cluster 7: scarce dependence between precipitation and elevation, but only 16 grid cells out of 96 have H > 500 m.
Comparisons between satellite products and GPCC_FD v6: the terrain elevation (3/3)
RMSE as a function of the elevation
Generally there is an increase in the RMSE
with the elevation.
3B42 has the lowest dependence between
the RMSE and the elevation and the
lowest RMSE values.
2000 3000 m
1000 2000 m
500 1000 m
200 500 m
0 200 m
Monthly mean anomalies with respect to GPCC_CLIM data set Satellite precipitation datasets
can be used to construct
short-term climatologies
exploited in regional studies as
regards the occurrence of
extreme events and their
connections with large-scale
climatic drivers, topography
and sea surface temperature.
cluster 1
cluster 3
The agreement is better for clusters
1, 3, 7, and 8, particularly in recent
years
Cluster 1 covers most of Kenya, one
of the EA countries mainly affected
by droughts together with Somalia
and Ethiopia.
drought events
TAMSAT
GSMAP
3B42
PERSIANN
RFE
CMORPH
3B31
GPCC_FD
Monthly mean anomalies with respect to GPCC_CLIM data set
TAMSAT
GSMAP
3B42
PERSIANN
RFE
CMORPH
3B31
GPCC_FD
cluster 2
cluster 4
For clusters 2, 4, 5, and 6
the agreement among
the anomalies worse
except for 3B42.
Conclusions
The different precipitation annual cycles characteristics of the region are identified by
using the climatological data set GPCC_CLIM.
The satellite data sets correctly reproduce the annual cycle identified by means of
climatological data in terms of wet season duration, prevailing wet season (for bi-modal
cycles), and intensity peak months.
Insights on the satellite precipitation estimate variability can be obtained from the analysis
of the standard deviation of the six satellite product ensemble. The greater standard
deviation values are associated with mountainous areas and more intense precipitation.
Performances of satellite products were quantified by comparisons with GPCC_FD data set.
As expected and due to the use of GPCC data, 3B42 is the best performance satellite
product with MAE in the range 5-25 mm/month and best EFF 0.7 for all clusters.
CMORPH and RFE make also use of rain gauge estimates for precipitation bias correction,
but their agreement with GPCC_FD data is not as good as that of 3B42. Nevertheless both
products have positive EFF.
RFE and TAMSAT give similar results in terms of MAE, RMSE, and EFF (≥ 0.5) and
lower BIAS for TAMSAT. The TAMSAT dry BIAS is recognized in Maidment et al. (2014)
and attributed to the approach used in algorithm calibration, more oriented to
drought monitoring and low intensity precipitation retrieval.
TAMSAT is based on IR TB from Meteosat with a calibration methodology exploiting
historic rain gauge data across large climatically homogeneous regions. This
calibration methodology represents a valid alternative to the use of local satellite-
contemporaneous rain gauges, especially in EA with sparse gauge network.
GSMaP statistical scores are similar to those of CMORPH with BIAS values more
variable for GSMaP and better EFF score for CMORPH (cluster 5, 6, and 7). GSMaP
exploits a morphing approach of PMW rainfall similarly to CMORPH, but unlike
CMORPH it does not include the bias correction by means of rain gauge data.
Orography represents an issue for the precipitation estimate from satellite, this is
evident from the analysis of the six member ensemble standard deviation and the
increase of the RMSE as a function of the terrain elevation.
Article
Full-text available
This paper presents an extensive validation of the combined infrared/microwave H-SAF (EUMETSAT Satellite Application Facility on Support to Operational Hydrology and Water Management) precipitation product H03, for a 1-year period, using gauge observations from a relatively dense network of 233 stations over Greece. First, the quality of the interpolated data used to validate the precipitation product is assessed and a quality index is constructed based on parameters such as the density of the station network and the orography. Then, a validation analysis is conducted based on comparisons of satellite (H03) with interpolated rain gauge data to produce continuous and multi-categorical statistics at monthly and annual timescales by taking into account the different geophysical characteristics of the terrain (land, coast, sea, elevation). Finally, the impact of the quality of interpolated data on the validation statistics is examined in terms of different configurations of the interpolation model and the rain gauge network characteristics used in the interpolation. The possibility of using a quality index of the interpolated data as a filter in the validation procedure is also investigated. The continuous validation statistics show yearly root mean squared error (RMSE) and mean absolute error (MAE) corresponding to the 225 and 105 % of the mean rain rate, respectively. Mean error (ME) indicates a slight overall tendency for underestimation of the rain gauge rates, which takes large values for the high rain rates. In general, the H03 algorithm cannot retrieve very well the light (< 1 mm/h) and the convective type (>10 mm/h) precipitation. The poor correlation between satellite and gauge data points to algorithm problems in co-locating precipitation patterns. Seasonal comparison shows that retrieval errors are lower for cold months than in the summer months of the year. The multi-categorical statistics indicate that the H03 algorithm is able to discriminate efficiently the rain from the no rain events although a large number of rain events are missed. The most prominent feature is the very high false alarm ratio (FAR) (more than 70 %), the relatively low probability of detection (POD) (less than 40 %), and the overestimation of the rainy pixels. Although the different geophysical features of the terrain (land, coast, sea, elevation) and the quality of the interpolated data have an effect on the validation statistics, this, in general, is not significant and seems to be more distinct in the categorical than in the continuous statistics.
ResearchGate has not been able to resolve any references for this publication.